Nvidia announced Project DIGITS on Monday. It is a small desktop PC aimed at researchers and data scientists who want to experiment AI models, such as chatbots, like ChatGPT, and image generators, at home. The $3,000 device, containing Nvidia’s GB10 Grace Blackwell Superchip debuted at CES in Las Vegas. It will be available in May. It can work as a standalone computer or connect to Windows or Mac machines.
Nvidia CEO Jensen Huang presented the new system at CES on Monday. “a cloud computing platform that sits on your desk.” Project DIGITS was also designed as a bridge to connect desktop development with cloud deployment. Developers can test and create AI applications on Project DIGITS before moving them to cloud services, or data centers using similar Nvidia hardware.
The GB10 chip in the Project DIGITS computer combines a Nvidia Blackwell graphics card with a 20 core Grace CPU based upon Arm architecture. The chip was developed by Nvidia in partnership with MediaTek. It connects to up to 4TB storage and 128GB memory inside the Project DIGITS enclosure.
Running AI models locally
Many people are using AI models that require remote data centers to run due to their computational needs. Over time, there’s been a push to reduce the size of some AI models, so they can be run on locally owned hardware. Project DIGITS provides some of this capability at home.
According to reports, a single Project DIGITS can run AI models up to 200 billion variables. Two linked units can handle models of 405 billion parameters. The number of parameters in AI models roughly reflects the size and complexity of an AI model’s network. More parameters require more memory and computing power to run. Parameter size is also a good indicator of AI model capabilities, although different AI models perform differently based on how they are trained and architected.
